datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hallu_autod在使用框架之前，需要先准备以下数据集文件：
git clone https://huggingface.co/datasets/S3IC/hallu_autod
1. query_all.json
将 query_all.json 文件下载到以下目录：
OmniTrust/modules/hallucination/vlm/vlm_autodetect/query/query_all.json
2. 图像文件
根据 query_all.json 中的图像路径信息，将图像文件下载到对应的目录。默认情况下，图像文件应存放在：
OmniTrust/data/dataset/halu_autodetect/image/
注意：query_all.json 中的 filename 字段指向的是相对于项目根目录的路径，例如：
{
"id": 1,
"filename": "data/dataset/halu_autodetect/image/AMBER_1.jpg",
"question": "Describe this… See the full description on the dataset page: https://huggingface.co/datasets/S3IC/hallu_autod.Autoautolaparo-qna-sample
AutoLaparo QnA Sample (spec-aligned subtypes)
A small, spec-aligned QnA dataset for surgical-video VQA.
Each record follows the same schema as the production file output/qna_dataset.json,
and each question_subtype is one of the subtypes defined in the project's spec/*.md.
Built from AutoLaparo Task 1 (workflow phase recognition, video 14) and Task 3
(instrument and anatomy segmentation).
11 records / 11 subtypes — exactly one record per supported subtype to keep the
sample… See the full description on the dataset page: https://huggingface.co/datasets/phamdt2003/autolaparo-qna-sample.